1. CERN 21 January 2005Piotr Nyczyk, CERN1 R-GMA Basics and key concepts Monitoring framework for computing Grids – developed by EGEE-JRA1-UK, currently used by EGEE/LCG2 and experiment apps Based on GGF GMA definition – Producer-Consumer architecture But! With relational data model – the whole system appears as one large relational database Data is propagated from Producers to Consumers on many different levels and locations: core centers, regional centers (ROC), resource centers (RC) Central Registry is used to locate the data (producers) automatically. Webservice interface (servlets) However, it is not a general distributed RDBMS!

2. CERN 21 January 2005Piotr Nyczyk, CERN2 Types of queries and producers R-GMA supports three type of queries (extension to SQL syntax): CONTINUOUS SELECT – all new data as it is published LATEST SELECT – last value for given key(s) according to timestamps HISTORY SELECT – all historical data Corresponding Producers types: StreamProducer – published data is “broadcasted” and lives for certain short period, answers CONTINUOUS queries. LatestProducer – stores only last value for each key DBProducer – stores all published tuples DBProducers and LatestProducers are using physical DBMS to store the data (MySQL, others?)

3. CERN 21 January 2005Piotr Nyczyk, CERN3 R-GMA in EGEE/LCG Grid monitoring Motivation: We have a lot of monitoring tools (sensors) at different levels: tests running centrally, agents running on different sites, accounting information coming from Resource Brokers and sites … Data must be easy accessible from a single point (reports) Data must be archived but not in one place! There are lots of monitoring frameworks, but we already have R-GMA infrastructure in EGEE/LCG2 Decision: „we will use R-GMA as a central bus to distribute data between sensors and reporting tools”

4. CERN 21 January 2005Piotr Nyczyk, CERN4 Archivers (Secondary Producers) Archiver is a Consumer- Producer pair Consumer part is responsible for “catching” new data that is published by StreamProducers Producer part stores the data into DBMS (MySQL) and later can answer queries (Latest, History) Multiple archivers for fault tolerance (each has his own copy of data) DBProducer Archiver DBMS LatestProducer Archiver StreamProducer DBMS R-GMA

5. CERN 21 January 2005Piotr Nyczyk, CERN5 R-GMA as “Central monitoring bus”

6. CERN 21 January 2005Piotr Nyczyk, CERN6 MySQL based Archivers Currently we use MySQL as a backend for Archivers A single set of archivers contains two archivers: Archiver+LatestProducer, Archiver+DBProducer Two sets of Archivers for fault tolerance: CERN, Taipei Data survives archiver failure/restart as it is stored physically in external MySQL DBMS Insert rate: ~20M tuples/month, measurements each 5 seconds

7. CERN 21 January 2005Piotr Nyczyk, CERN7 Reporting Tool Prototype

8. CERN 21 January 2005Piotr Nyczyk, CERN8 Reporting Tool Prototype

9. CERN 21 January 2005Piotr Nyczyk, CERN9 Advantages No need for configuration of access points both for sensors and for reporting tools – data is located automatically by the registry No risk of data loss in case of failure of monitoring software – R-GMA is just for data transport, not for storage Flexibility – by using predicates, one can setup a number of archivers with different policies in different physical places, fault tolerance Usage of SQL and relational database model makes it all elegant

10. CERN 21 January 2005Piotr Nyczyk, CERN10 Known issues R-GMA Registry (and Schema) is a single point of failure – will be fixed in next release Supported subset of SQL sometimes not sufficient (aggregate functions, GROUP BY, etc.) Lack of schema modifications – once table is defined it can’t be changed or removed Performance issues: eg. lack of DB indices – however this can be done manually using direct access to DBMS Support for other DBMS: Oracle?

11. CERN 21 January 2005Piotr Nyczyk, CERN11 References R-GMA Home Page: http://www.r-gma.org/ EGEE-JRA1-UK Home Page: http://hepunx.rl.ac.uk/egee/jra1-uk/ R-GMA based monitoring system for EGEE/LCG2 operations: http://goc.grid.sinica.edu.tw/gocwiki/ RgmaUnifiedMonitoringSystem